Compensation MOSFETs (metal-oxide-semiconductor field-effect transistors) have p-type regions and n-type regions arranged next to each other under the actual device in the active volume in such a way that these regions can mutually ‘electrically compensate’ for each other during a blocking mode, so that a non-interrupted, low-resistance conductive path is formed in the switched-on state from the source to the drain. Each of the charging regions, also referred to as columns, contain only a fraction of the breakthrough surface charge qc when seen in the horizontal direction. Due to the special type of structure in this case, with a given breakdown voltage, the doping of the n-path can be massively increased for the structural components used for charge compensation compared with conventional structural components. A desirable reduction of switching resistance is achieved in this manner and a distinctly increased added value is therefore also achieved.
However, one part of this added value is typically negated by the increased expense incurred to create the complicated p-type and n-type column regions of the voltage absorbing volume. Conventional manufacturing methods used to create such charge compensation structures are quite complicated and have very high costs, and therefore greatly reduce the added value that is achieved with the compensation structures. In addition, the lower limit of possible dimensions of the charge compensation structures available with existing techniques has been reached, so that further shrinking in size will be difficult to achieve in the next technological generations. A technological transformation has not been realized so far because the required manufacturing tolerances could not be achieved with conventional processes.
Charge compensation components are mainly produced with a so-called ‘multiple epitaxy’ process. In this case, an n-doped epitaxial layer, which is several μm thick, is first grown on a highly n-doped substrate and commonly referred to as ‘buffer epi’. In addition to a doping level introduced in the epitaxial step doping ions are introduced into the buffer epi through a photoresist mask using implantation with the doping ions in the first charging locations (for example boron for phosphorous doping). Counter doping can be also employed with implantation (either through a mask, or on the entire surface). However, it is also possible to separate the individual epitaxial layers with the required doping. After that, the entire process is repeated as much time as required until an n (multi-epitaxial) layer is created which has a sufficient thickness and which is equipped with charge centers. The charge centers are mutually adjusted to each other and vertically stacked on top of each other. These centers are then merged with outward thermal diffusion in an undulating, vertical column to form adjacent p-type and n-type charge compensation regions. The manufacturing of the actual devices can then be conducted at this point. Due to the fact that several expensive epitaxial steps are used including intermediate operations such as photo technology, implantation, etc., the ‘multiple epitaxy’ process explained above is very expensive and time consuming.
Another conventional technique for fabricating charge compensation components involves trench etching and compensation with trench filling. Initial development of this process included discussions during which both column types (mostly the p-type columns) were defined with trench etching, followed by epitaxial filling to prevent disadvantageous costs which arise when using several different epitaxial layers as explained above with regard to the ‘multiple epitaxy’ process. At the same time, the volume which absorbs the voltage is deposited in a single epitaxial step (n-doped epi) on a highly n-doped substrate, so that the thickness corresponds to the total thickness of the multilayered epitaxial structure. After that, a deeper trench is etched, which determines the form of the p-column. This trench is then filled with p-doped epi which is free of crystal defects. Indeed, the charging, conditioning and cleaning steps are identical and the actual duration of the process is not the main factor driving the higher cost. On the other hand, the ‘multiple epitaxy’ process provides savings during the column photo technology steps and results in savings related to many deposits for the epi layers of the columns. Also, with multiple epitaxy, the implantation centers which are stacked on top of each other must be vertically merged together with thermal diffusion.
At the same time, the regions are diffused in the same manner also laterally, which leads to a distinct limit that is imposed on the minimum width of the structure. When the columns are doped already during the initial epi process, just like with the trench filling method, there is no need for such a diffusion step, so that a structure which has very narrow columns can be produced and a higher cost-performance ratio can be achieved. However, the integration of doping by using an epi process is possible only with relatively large fluctuations. In particular with very small dimensions, the corresponding fluctuations quickly exceed the window provided for the process, which can lead to significant yield losses. Also, a vertical variation of the doping profile (and thus also of the vertical development of the strength of the field) is not possible. Various robustness criteria therefore cannot be met.
Another conventional technique for fabricating charge compensation components involves co-doping of the starting material with different quickly diffusing doping atoms followed by trench etching and intrinsic epi filling. However, the problems associated with precise doping are circumvented so that the trench geometry does not change the charging balance. A new characteristic, when compared to the manufacturing concepts discussed so far, is the doping of the epi starting layer. In particular, the epi volume contains both elements Ep (p-type) and En (n-type), which later form the p-type and n-type compensation columns in the finished product, although the columns are spatially separated from each other. This ‘double doping’ can be produced by simultaneously adjusting both of the doping gas currents during the epi growth or so that the entire gas volume consists of a multi-epitaxial sequence, where the doping implantation is not masked and instead both of the doping types are incorporated on the entire surface with the desired doses.
The horizontal levels are merged in the vertical direction with a strong outward diffusion into a continuous (undulating) doped material. Immediately after the starting epitaxy, while both doping materials are not yet spatially separated from each other, they are homogenously distributed in each (intrinsically thin) horizontal layer and mutually compensate for each other or a doping gradient can be built into each of both elements in the vertical direction. As already mentioned, opposite doping polarity must be realized (meaning that one element must have the p-doping effect and the other one must have the n-doping effect. The diffusion coefficients of both doping elements in silicon must be very different with customary diffusion temperatures of around 1,000° C. to 1,250° C. An example of such a doping pair is As (n-doping) and B (p-doping). The diffusion speed of boron is in the given temperature range about three times as high as that of arsenic.
A lateral separation of the doping element in the charge concentration points can be achieved so that the desired doping columns are created by etching a deep trench in the starting epi (which is doped with both doping elements Ep and En) so that a mesa structure still remains in front of the starting epi. Doing so conserves the horizontal charge balance. An undoped epi layer is then grown on the lateral walls of the trench. This also has no influence on the lateral charge balance. The trench can then be filled at this point. A strong outward diffusion of the doping elements Ep and En is carried out after that or at a later point during the process (for example during the device process). Since the diffusion coefficients of both doping elements are very different, a large amount of the doping substance of the faster diffusing element diffuses into the undoped epi layer covering the trench sidewalls. A disproportionally higher ratio of the more slowly diffusing element is diffused into the remaining silicon (Si) mesa layer. One part of the doping amounts is compensated for intrinsically. This occurs in particular in the remaining Si mesa layer. The non-intrinsically compensated doping substance amounts are electrically active, and are determined for each location by the difference in the concentrations of Ep and En. This effect also substantially builds the doping column. It follows from the description above that with the homogenous doping of the starting epi, no individual processes including etching of the trench, trench wall depositing with epi and column diffusion change the charge balance relative to the starting status of the double-doped starting epi.
The starting epi co-doping/trench etching/intrinsic epi filling technique yields compensation columns with small dimensions, so that the entire amount of the charge can be used to control with precision the voltage increasing volume. However, with this technique, the n-doping atoms never diffuse in silicon faster than the p-doping elements. The charge separation thus occurs mostly because the Ep atoms are diffused out of the pre-deposed region of the Ep atoms, while the En profile is hardly changed at all and therefore is only made slightly fluid. This results in several problems.
First, a higher ratio of p-doping atoms remains in the n-type columns, and although they are intrinsically compensated for with an oversupply of En, the effect on the silicon grid reduces the electric mobility of the electrons (there are about three times as many doping elements in the n column as there are in the p column where no intrinsic compensation takes place). Therefore, although the breakdown voltage is increased in this manner, since the n column carries the load current in the switched-on status of the transistor, this characteristic increases also the switched-on resistance.
Furthermore, the Ep atoms diffuse faster also in the vertical direction than the En atoms, creating a more highly doped p-type region under the n-type column, unless a highly doped substrate layer is located under the column. The n-type column should be coupled so that it is conductive with n-conductive compensation components in the downward direction to the rear side of the device. However, this cannot be achieved (due to the p-layer described above) without additional measures.
Also, in order to ensure the robustness of the compensation component, methods were used which were based on the fact that the amounts of the doping substance were varied in the vertical direction in the p-type column and/or in the n-type column. The goal is to produce an electric peak with about a half of the height of the voltage absorbing volume. For precision reasons, the simplest way to achieve this is when the starting epi layer is manufactured with a multilayered epitaxy. The doping is performed with implantation of En and Ep on the entire surface in each individual epi level, and the implantation dose can be varied. With a subsequent diffusion (which occurs still before the etching of the trench), the doping substance is vertically distributed throughout the epi volume and therefore throughout the individual epi regions. In this case, the En diffusion is much slower than that of Ep. When the En doped starting epi is diffused at the same time in the vertical direction, the diffusion must be conducted until the doping elements are homogeneously distributed mostly vertically through the epi volume. A doping profile, which is predetermined with the implantation dose for the individual epi level, is therefore not retained (or only an augmented one may remain). Due to these occurrences, it is very difficult to build a vertical doping profile having a high electric robustness.
There is also the fact that a higher percentage of the Ep atoms are diffused in the vertical direction from the voltage absorbing volume, for example into the low laying substrate. The vertical outward diffusion of the Ep atoms interferes with the horizontal charge balance if no measures are taken against the vertical outward diffusion, such as for example buried oxide layers. The vertical outward diffusion of the Ep atoms also has a detrimental influence on the finishing tolerances.
Another conventional technique for fabricating charge compensation components involves doping of the trench sidewalls with implantation. That is, defining the p-type compensation columns can be formed with implantation instead of filling the trench. The trench itself can then be filled with a dielectric, or even remain unfilled and then closed up only in the upward direction. However, this technique results in reflection mechanisms at the sidewall of the trench which plays an important role. Further, the location in which the doping substance is inserted depends to a great deal on the angle of implantation and on the geometry of the trench (the window available for the implantation is not sufficiently large). Moreover, this technique also does not make it possible to vary the charge balance or the relationships between the fields which are related to the depth even though this option is important for charge compensation components, namely in order to ensure the current capability with a full load. Similar performance characteristics are relevant in so-called avalanche switch occurrences.
Another conventional technique for fabricating charge compensation components involves implantation with ultra high energy. The areas that are based on implantation using extremely high energy amounts in theory capture the increasing voltage volume in a single epi step in the depth of the layer. The p-type columns should then be defined by implantation in several stages using different energy levels through a thick surface mask. The required column depth should be achieved with the very high implantation energy level. However, attempts in these areas have not been successful because a suitable masking process is not available.